workplace-based vocational training in central and eastern europe - evidence on the school to work...

8/12/2019 Workplace-Based Vocational Training in Central and Eastern Europe - Evidence on the School to Work Transition of Hungarian Apprentices

1/36

Workplace-Based Vocational Training in Central and Eastern Europe - Evidence on the School to

Work Transition of Hungarian Apprentices

Daniel Horn


2/36


3/36

Daniel Horn

Center for Economic and Regional Studies, HAS and ELTEcon

WP4 Task 3P.4.8.

Versionfinal 2013. 10. 29.

Workplace-Based Vocational Training in Central and Eastern Europe - Evidence on the School to

Work Transition of Hungarian Apprentices1

Abstract:

Although apprenticeship training has been praised for its effectiveness in smoothing the school-

to-work transition of non-college bound students in Western European dual education systems,

there is a lack of evidence from Central Eastern Europe. Using a unique individual-level panel

database, which includes an extensive set of controls, the study shows that Hungarian students of

the non-college bound vocational training track with workplace-based training, have about 10-15%

higher probability of initial employment, compared to similar graduates from the same track, who

were trained in school. This effect seems to be stable across industries, and robust to specification

checks.


4/36

1. Introduction2There is a serious lack of studies on the effectiveness of workplace-based vocational training in

Central-Eastern Europe (but see Baranowska, Gebel, and Kotowska 2011; Noelke and Horn 2011).

Although, it has been widely argued that workplace-based training helps preparing non-college

bound youth for the labor market, the arguments are almost solely based on studies from Western

Europe, mostly from dual vocational education and training (VET) systems. These studies show that

secondary level education systems, combining school-based vocational education with employer-

provided, workplace-based (apprentice) training, have sustained a positive track record in smoothing

the school to work transition process, lowering the unemployment rate, and increasing the quality of

work (Rosenbaum et al. 1990; Mller and Shavit 1998; Shavit and Mller 2000; Ryan 2001; Breen

2005; Wolbers 2007; Wolter and Ryan 2011; Piopiunik and Ryan 2012). Besides the lack of empirics

from Central-Eastern Europe, existing empirical research provides little information about the causaleffect of apprentice training on school-to-work transition.

The paper improves on the existing literature in these aspects. Firstly, it provides information on the

effectiveness of apprentice training from a non-dual education system in a Central-Eastern European

country, Hungary. Secondly, there has been only a handful of studies that tried to show the causal

effect of workplace-based training on labor market outcomes (Bonnal, Mendes, and Sofer 2002;


5/36

socialist era as well, but its essence has changed dramatically, and has much less resemblance today

to the traditional dual systems.

The nature of apprenticeship training in socialist countries was changed mainly due to the huge

national enterprises, as compared to the pre-war network of smaller independent firms and

businesses, where apprenticeship training had originated. The provision as well as the content of

training changed markedly. Also since the competitive market was demolished, and people were

essentially guaranteed a job, having an early work-experience mattered much less both for the

individual and the firm, which decreased the motivational aspects for both parties (see West 2013 for

a comprehensive review of the transition of the CEE VET systems). Under the socialist era the

students spent most their practical training time in large enterprises, rather than being in contract

with several smaller factories, but at the same time not being jointly educated in shared off-the-job

facilities, as in the dual apprentice training systems (Kogan 2008)

Moreover, although the post-transition changes in these countries have seen the demolishment of

the national enterprises and the revival of small businesses, the development of the education

system was highly decentralized (Horn 2010), and thus the central coordination of vocational

workplace-based training between business and the state was secondary, at best. In the beginning

of the privatization and restructuring processes in CEE countries [], employers largely withdrew


6/36

that look at Central and Eastern Europe. Also getting comparable statistics is difficult due to the

different understanding of apprenticeship training: many CEE countries refer to their vocational

training programs as apprentice training programs, although students dont have any contractual

agreements with firms (Kogan 2008, 2122), or label non-formal forms of training as apprentice

training3.

To my best knowledge there have been only two comprehensive reviews on the vocational education

sector in Central and Eastern Europe. Kogan, Gebel and Noelke (2008) have relied on local expertise

for their handbook to describe the general education system, including its VET, of the given

country. With this local help they have categorized the countries into school-based or dual VET

(see table 1 below).

Table 1. The predominant form of organization of VET

School-based Dual system

Bulgaria X

Czech Republic X

Estonia X

Hungary X

Latvia X

Lith i X


7/36

52). Thus it seems from these two sources that apprenticeship training is still significant in the

previous Austro-Hungarian empire countries and in Poland (with strong linkages to Germany), even

though apprenticeship training (or dual-education) might not mean the same as in Western Europe.

Looking at official statistics is less helpful due to the lack of comparable education data on

apprenticeship training, and due to the different understanding of the term. On the other hand, one

might approach training from the other direction, from the business sector. Table 2 below shows the

training enterprises as percentage of all enterprises. Apparently the typical dual education systems of

Austria, Denmark or Germany are all above the EU mean. But the conclusion from this table is less

straightforward. Firstly, this statistic is also high in non-dual Western countries (as the UK or Sweden

and Finland), and secondly, countries from Central and Eastern Europe are scattered along the

distribution (although on average below the EU mean). Note, however, that with the exception of

Estonia all ex-Austro-Hungarian empire members have higher percentages of training firms than

the other CEE countries. This also points toward the fact that workplace-based training can be

important in the CEE countries, and especially in those, where dual-training have a tradition. But

whether apprentice training is just as effective in this region as it is in the Western dual economies is

yet unclear.


8/36

Table 2. Training enterprises as % of all enterprises

2005 2010

United Kingdom 90 80

Norway 86 n.a.

Denmark 85 n.a.

Austria 81 87

Sweden 78 87

Finland 77 74

Netherlands 75 79

France 74 76

Slovenia 73 68

Czech Republic 72 72

Luxembourg 72 71

Germany (until 1990 former territory of the FRG) 69 73

Estonia 67 68

Ireland 67 n.a.

Belgium 63 78Slovakia 60 69

European Union (27 countries) 60 66

Croatia n.a. 57

Cyprus 51 72

Hungary 49 49

Spain 47 75

Lith i 46 52


9/36

In Hungary every student in the lowest non-college bound vocational training track had to do at

least two years of practical training, which could either be done in the school or at a firm (see the

description of the Hungarian VET system below).5This study compares these two groups of students.

Hence the treatment and the control groups are quite obvious: both groups have received

exactly the same general training (the first two years in the vocational training program), and

provided they aim for the same occupational qualification they have to fulfill the same

requirements; the only difference between the groups is the place where practical training takes

place. Although the allocation of students between training places might not be random, the HLCS

offers an exceptionally wide variety of individual controls, which reduces the omitted variable bias

concern. The database also includes information on the types of qualification that students have

acquired, and on the schools of the students, which allows for within industry and within school

analysis, that further strengthens the reliability of the estimates. Moreover, the HLCS is a panel

database, which rules out the problem of reverse causality.

4. Previous research on causal effectsThere are but a handful of empirical studies that offer analysis of the causal effects of apprenticeship

t i i i di id l l l l b k t t ( i b W lt d R 2011) Th


10/36

The authors argue that this finding could be due to three distinct reasons, among which they are

unable to discriminate: a) apprentices might lack the general human capital, as opposed to non-

apprentice VET students, and thus finding a job at a firm other than their training firm is harder/not-

easier; b) movers might be negatively selected, as those who are not hired by the training firm

might have some unobserved negative trait; and similarly c) there might be a negative signaling

effect associated with moving to another firm, even if movers are not different from stayers in

other respects.

A similarly designed study is Bertschy, Cattaneo and Wolter (2009), who look at the Swiss dual

system. They also use a panel which is connected to the PISA 2000 Swiss database, which provides

standardized test scores to proxy student achievement as well as socio-economic status and other

controls, and they also use simultaneously estimated equations to take selection into account.

However, since the vast majority of the Swiss vocational students (over 90%) are in the dual

apprenticeship training, they compare apprentices, who taken up training with higher intellectual

level, with the others. Also their utilized outcome is not employment, but employment in adequate

job that matches the graduates qualifications. Initially they find a significant difference between

these two groups, which disappears after they take selection into tracks into account. They

emphasize that self-selection into educational tracks is very important. In fact, students with higher

PISA lit l lik l t d t d lik l t ll i ti l fi ld ith


11/36

5. A non-dual CEE system - the Hungarian VET systemWhile most of the studies that have addressed the question of the effectiveness of apprenticeship

training are based on countries with dual systems, the Hungarian VET also allows for a within track

comparison of workplace-based and school-based training. As a non-dual system this country study

should especially be important for countries with less experience in apprentice training. Thus findings

from a non-dual system, where the workplace-based training is still widely utilized, could be

informative for those countries where apprenticeship training is less widely spread but its

development is considered.

Also, as van de Werfhorst and co-authors (van de Werfhorst 2011b; Bol and van de Werfhorst 2011;

van de Werfhorst 2011a) have pointed out different theories might explain better the education-

labor market link in different countries or in different labor market settings. Specifically van de

Werfhorst (2011b) argues that in dual systems the human capital theory is more adequate, since

there is a stronger match between the skills acquired in education and skills needed on the

workplace. In countries with less evident link between labor market and education, other indicators

of skills such as general literacy or numeracy are more important, and thus educational

attainment is less important. Although in his comparative studies (van de Werfhorst 2011b; Bol and

van de Werfhorst 2011) Hungary is considered as a strongly vocationally oriented country, and as


12/36

repeaters, or those students with special educational needs (SEN), who cannot be integrated with

the others and thus enrolled in special vocational training. While both the academic and the

vocational secondary tracks offers general training for four years - and the vocational secondary

offers pre-vocational training, with usually one or two optional years of vocational practical training

after the school-leaving exam the vocational training track has so far offered only two years of

general training7with two additional years of practical training.

This paper focuses on the 20%, who are enrolled in the vocational training (VT) track. This track is

considered to be the lowest ranked in the hierarchy of tracks (but still above no-education). This

paper compares VT students who have done practical training at a private firm with those, who have

done practical training within schools.

Although the selection into workplace-based training might not be random, there is no central

procedure that allocates students in one group or in another. In fact, the organization of the system

is overly school-based, with relatively few links to the labor market (Kis et al. 2008). The system has

been one of the most decentralized ones in the OECD (OECD 2004). It is the duty of the school to

provide practical training for the student. The school can either organize the training within its

boundaries (e.g. by hiring vocational teachers) or can outsource the training to a private firm,

which can be done in groups or individually as well. The student can also organize training for


13/36

between workplace-based and school-based training. After the introduction of the HLCS data, I will

address this question.

6. The HLCS dataThe Hungarian Life Course Survey (HLCS) is an individual panel survey conducted annually. The

original sample of 10,022 respondents was chosen in 2006 from the population of 108,932 eighth

grade students with valid test scores from the National Assessment of Basic Competencies (NABC).

The NABC measures the literacy and numeracy of all 6th

, 8th

and 10th

grade students every year,

starting from 2006 (OECD 2010). The NABC also contains a set of family background variables, such

as parental education or employment status. The first HLCS survey wave was completed during the

winter of the school-year 2006/7, and subsequent waves have been fielded on a yearly basis.

Currently there are 6 waves available with fairly large response rates. The annual sample attrition

rate, on average, is only around 5% (see Table 3).

Table 3. Basic statistics of the HLCS database

wave School Date of

th

Median school grade Number of students (with

li SEN

Number of students

( t ti b


14/36

students with special educational needs (SEN) are overrepresented in the data, propensity weights

are used to control for the oversampling, as well as for the imminent sample attrition. The following

strata were used during the data collection, and in estimating the weights: 1) 3 settlement types: the

capital and big cities, other cities, villages 2) 7 NUTS-2 regions93) Reading literacy test scores (30

equal groups from the NABC 2006 reading literacy distribution).

The most important variables of interest in this paper are the school track, the apprenticeship status,

and the labor market outcome. School track is defined as the students school track in the 4 thwave of

the study, the year when the median student was finishing the last year of compulsory schooling. All

students in the analysis were enrolled in the vocational training track in the 4th

wave. Vocational

training students could either do their practical training within school in class, or in a school

workshop, or could go to a private firm, either with the help of the school (usually in groups), or by

organizing the training by themselves. I have labeled the former two as school-based and the latter

two as workplace-based training. Anyone, who did workplace-based training in the 4th

wave or in the

5th

wave of the study (the year after finishing compulsory education), is considered an apprentice.10

The four types of labor market outcomes employed, unemployed, studying and other11 are

considered in the last (available) wave of the study, and are self-declared. The main reason for using

the 6th

wave and not the 5th

wave, which is the one after the compulsory education ends, is that the

j i f d i h 5th ill i d i h i l i i


15/36

15

Table 4: Labor market outcomes in the 5th

and 6th

wave

5th wave 6th wave

work unempl. study other missing Total work unempl. study other missing Total

academic 70 54 1717 62 172 2075 187 95 1419 85 289 2075

% 3,37 2,6 82,75 2,99 8,29 100 9,01 4,58 68,39 4,1 13,93 100

voc.sec. 106 115 2037 62 158 2478 452 303 1219 161 343 2478

% 4,28 4,64 82,2 2,5 6,38 100 18,24 12,23 49,19 6,5 13,84 100

voc.tr. 148 189 958 62 114 1471 541 290 286 123 231 1471

% 10,06 12,85 65,13 4,21 7,75 100 36,78 19,71 19,44 8,36 15,7 100

spec.voc.tr. 23 34 191 12 26 286 60 45 108 25 48 286

% 8,04 11,89 66,78 4,2 9,09 100 20,98 15,73 37,76 8,74 16,78 100

missing 252 418 906 246 1890 3712 508 408 515 262 2019 3712% 6,79 11,26 24,41 6,63 50,92 100 13,69 10,99 13,87 7,06 54,39 100

Total 599 810 5809 444 2360 10022 1748 1141 3547 656 2930 10022

% 5,98 8,08 57,96 4,43 23,55 100 17,44 11,38 35,39 6,55 29,24 100


16/36

Other variables that are used are the standardized test score (literacy and numeracy) in 8th

grade,12

class mark averages (1- fail to 5- excellent) in 8th

and in 12th

grade, gender, SEN status, Roma

ethnicity, parental education and occupation. All control variables are from the first wave of the

study, unless otherwise noted. Additional controls are a proxy for grade repetition (whether the

student was in the 12th

grade in the 4th

wave of the study) and a proxy for motivation (whether

her/his 9th

grade school was her/his first choice). The size of the training firm (small: 1 to 12

employed, medium: 13 to 100 employed and large: over 100 employed) is also used in some

estimations. The month of survey is controlled in all estimations and is not shown. See Table 5 for

descriptive statistics.

Table 5: Descriptive statistics

data available for students in the 6

th

wave of HLCSVocational training students only

Variable obs. weighted obs. mean s.d. min. max.

apprentice, 4thor 5

thwave 1183 15048 0.60 0.49 0 1

math test score (std.), 8thgrade 1087 14180 -0.83 0.68 -2.74 2.10reading test score (std.), 8

tgrade 1217 15447 -0.92 0.68 -3.78 1.21

class mark average, 8tgrade 1170 14883 3.18 0.53 1 5

l k 12

th

d 1217 15447 3 32 0 58 2 5


17/36

students only 964 has this information in the dataset. The official list of OKJ qualifications contains 21

larger categories. I have grouped these into 6 broad categories (industries) in order to increase the

number of cases within each category, but still facilitate relevant comparison between the groups

(see Table A1. in the appendix).

Table 6.: Number and percentage of VT students in school- and workplace-based training by industry

Industry school-based work-based missing Total

social services 3 6 0 9

% 33,33 66,67 0 100

mechanics 108 112 4 224

% 48,21 50 1,79 100

industry 124 106 2 232

% 53,45 45,69 0,86 100

transport-environment 13 19 0 32

% 40,63 59,38 0 100

services 121 267 7 395

% 30,63 67,59 1,77 100

agriculture 43 29 0 72

% 59,72 40,28 0 100

missing 178 296 33 507

% 35,11 58,38 6,51 100


18/36

Covariates that are significant in the first estimation (Table 7 column 1) suggest that higher skilled

students are more likely to enter apprenticeship training. The within industry estimation (column 2)

as well as the within school estimation (column 3) do not show these strong skill differences between

apprentices and non-apprentices, suggesting that the (self)selection into occupations or more likely

into schools drives the results. That is, there might be some occupations and/or schools that attract

better students (see also Bertschy, Cattaneo, and Wolter 2009). The base (column 1) and the within

school estimations (column 3) also show that people with less educated parents are more likely to

have practical training at private firms. The results in the most restrictive within school and within

occupation model (column 4) however highlight that none of the individual traits matter if

occupational differences and school and/or local labor market effects are taken into account.

Individual traits explain only 4% of the variation of the base model, but industry effects add an

additional 4%, suggesting that students in different occupations have different chances of getting

into apprenticeship (see also table 6). The inclusion of school fixed effects in the model increases the

fit tremendously. The school fixed effect model explains is almost 50% of the variance, which is

further increased to 73% when industry fixed effects are added. This is of course not surprising, given

that the sample was not representative on the school level and thus there are on average less than 4

students per school in the sample, which further decreases when within industry effects are

id d i hi h l


19/36

Table 7: Selection into apprenticeship linear probability models

(1) (2) (3) (4)

class mark average, 8th grade 0.0914** 0.0321 0.0289 -0.0428

(0.0413) (0.0505) (0.0444) (0.0794)

class mark average, 12th grade -0.0124 -0.0118 0.0611 0.0925

(0.0334) (0.0396) (0.0377) (0.0626)

math test score (std.), 8th grade -0.0141 -0.00424 -0.0280 -0.0182

(0.0318) (0.0378) (0.0318) (0.0530)

reading test score (std.), 8th grade 0.0624** 0.0337 0.0263 -0.0380

(0.0308) (0.0406) (0.0338) (0.0652)parents' ed.: primary or below 0.0866* 0.0689 0.134*** 0.0927

(0.0482) (0.0625) (0.0513) (0.0912)

parents' ed.: secondary or higher 0.0151 0.0260 0.00147 0.0226

(0.0437) (0.0521) (0.0456) (0.0782)

father employed, 4th wave -0.0553 -0.0292 -0.0711 -0.105

(0.0437) (0.0526) (0.0487) (0.0877)

father unemployed, 4th wave -0.0270 -0.00995 -0.0469 0.00422

(0.0508) (0.0667) (0.0539) (0.0911)

SEN student 0.0294 0.0210 0.0795 0.0878

(0.103) (0.0935) (0.103) (0.172)

Roma -0.0429 -0.0581 -0.0223 0.103

(0.0647) (0.0862) (0.0655) (0.119)

9th grade track is first choice 0.0323 0.0775 0.0531 0.104

(0.0426) (0.0549) (0.0441) (0.0743)

12th grader in 4th wave 0.142*** 0.151*** 0.0481 0.0822

(0.0476) (0.0547) (0.0483) (0.0768)

f l 0 0159 0 0680 0 0230 0 00427


20/36

significant in all three estimations, and show, that those VT students who had carried out practical

training at a private firm, as opposed to doing practical training in school, have around 1.5-1.6 times

higher odds of being employed, as opposed to being unemployed. The size of this effect is

unchanged by any of the personal traits that are included in the model. On the other hand social

background, gender and grade repetition matters in getting a job. It seems that students with

employed fathers have much higher odds of being employed; whether this effect materializes

through socialization or though social networks is not obvious. Also men are more likely to be

employed and women are more likely to fall into the other category (e.g. maternity leave). Students,

who have not repeated grades until 12th

grade, are also more likely to be employed in the 6th

wave of

the study. Note, however, that none of the school achievement variables neither the standardized

test scores, nor the teacher given class marks seem to be relevant in employment, although

students with higher class marks are more likely to study than to be unemployed.

Table 9 adds further controls to the base model. Table 9 block 1 is the same as Table 8 block 3 to

facilitate comparison of models. Table 9 block 2 shows the same multinomial logit model with

industry fixed effects added,14

while dummies of the training firm size are used in block 3 instead of

the apprenticeship dummy. The main conclusion does not change even if these controls are added:

apprentices have significantly higher odds to be employed vs. being unemployed than those with

l h l b d l l h h h ff l h l


21/36

Table 8: Effects of apprenticeship training, base model - multinomial logit model, odds of being employed, studying or other wrt. being unemployed

(1) (2) (3the base model)

work study other work study other work study other

apprentice, 4thor 5thwave 1.489** 1.149 1.226 1.457* 0.975 1.179 1.648** 1.106 1.211

(0.264) (0.232) (0.319) (0.283) (0.216) (0.324) (0.335) (0.253) (0.401)

class mark average, 8th grade 1.186 1.367 1.373 1.210 1.347 1.384

(0.258) (0.325) (0.438) (0.265) (0.312) (0.534)

class mark average, 12th grade 1.136 1.636** 1.597* 1.056 1.618** 1.344

(0.197) (0.323) (0.386) (0.188) (0.320) (0.436)

math test score (std.), 8th grade 1.170 1.062 0.720 0.963 0.972 0.974

(0.221) (0.216) (0.184) (0.184) (0.203) (0.272)

reading test score (std.), 8th grade 0.724* 1.023 0.897 0.797 1.064 0.681

(0.128) (0.209) (0.233) (0.144) (0.217) (0.203)

parents' ed.: primary or below 0.624* 0.529** 0.789

(0.167) (0.157) (0.317)

parents' ed.: secondary or higher 0.985 1.243 1.534

(0.235) (0.341) (0.712)

father employed, 4th wave 1.841** 1.247 1.707(0.443) (0.343) (0.693)

father unemployed, 4th wave 0.927 0.685 0.839

(0.254) (0.216) (0.365)

SEN student 0.807 0.851 8.19e-08***

(0.413) (0.490) (5.52e-08)

Roma 0.877 1.071 3.538***

(0.289) (0.433) (1.525)

9th grade track is first choice 1.015 1.017 1.084

(0.231) (0.259) (0.406)

12th grader in 4th wave 1.851** 0.603* 0.730

(0.483) (0.161) (0.279)

female 0.539*** 0.987 10.18***

(0.124) (0.257) (4.456)

Constant 1.487*** 0.923 0.349*** 0.501 0.0786*** 0.0190*** 0.372 0.154* 0.00588***

(0.193) (0.137) (0.0640) (0.415) (0.0754) (0.0207) (0.328) (0.158) (0.00820)

Observations 1,183 1,183 1,183 1,025 1,025 1,025 964 964 964

Standard error in parentheses, *** p


22/36

Table 9: Effects of apprenticeship training, industry FE - multinomial logit model, odds of being employed, studying or other wrt. being unemployed

(1the base model) (2) (3)

VARIABLES work study other work study other work study other

apprentice, 4th

or 5th

wave 1.648** 1.106 1.211 1.826** 0.978 1.329

(0.335) (0.253) (0.401) (0.479) (0.275) (0.594)

apprentice firm size, small (1-12)+ 1.496 0.790 0.983

(0.477) (0.276) (0.637)

apprentice firm size, medium (13-100)+ 3.926*** 2.275* 5.083**

(1.663) (1.039) (3.246)

apprentice firm size, large (100+)+ 1.703 0.937 0.799

(0.597) (0.381) (0.573)

Constant 0.372 0.154* 0.00588*** 0.103 0.563 0.0572 0.0739* 0.457 0.0450

(0.328) (0.158) (0.00820) (0.153) (0.863) (0.148) (0.112) (0.705) (0.109)

Industry FE no no no yes yes yes yes yes yes

Observations 964 964 964 679 679 679 670 670 670

Standard error in parentheses, *** p


23/36

Table 10 shows the predicted probabilities and marginal effects of apprenticeship training. The

baseline uncontrolled average probability of being employed for a VT student in 2011 is 44%.

Apprentices, however have a 47.1% chance, while school-based trained students have a 39.6%

chance of being employed. The chances of being unemployed is the reverse: apprentices have a 21%

chance, while the others have a 26.5% chance. There are no differences in the uncontrolled average

baseline probabilities of the other two outcomes between the two groups (study: 24%, other: 9%).

Using the base model (Table 8 block 3) to predict the probabilities at the population means similar

but somewhat higher percentages are gained. The predicted a probability of being employed for

apprentices is 52.3%, while for the school-trained it is 41.4%; the marginal effect of being trained at a

private firm is thus 10.9 % at the mean. In other words, the average apprentice has about 11% higher

chance of being employed after graduation than the average non-apprentice. This effect is somewhat

lower for the top of the distribution students (high class mark averages, high literacy and numeracy,

and parents with secondary general or tertiary schooling) and higher for the lower status lower

skilled students (low class mark averages, low literacy and numeracy, and parents education primary

or below). While the former group has 8.7% higher probability of being employed the latter has

11.5%. The marginal effects are also larger for apprentices, who were trained in mid-sized firms. The

average marginal effect here is 19%, but bottom of the distribution students benefit more (25.7%)


24/36

Table 10: Predicted probabilities and marginal effects

Predicted probability Marginal effect

school-based

training

workplace-based

training

workplace-based training

at population mean atpopulati

on mean

For a lowachiever low

status

student*

For a highachiever, high

status

student**

uncontrolled 39.6 47.1 - - -

base model 41.4 52.3 10.9 11.5 8.7

with industry FE 35.6 50.4 14.8 14.9 13.4

at a medium size firm 34.9 53.9 19.0 25.7 14.5within industry

at population mean at population

mean

at industry mean

social services 13.9 23.0 9.4 14.6

mechanics 39.8 54.9 15.2 14.8

industry 36.7 51.6 15.0 15.2

transport-environment 44.9 60.3 15.3 15.3

services 33.7 48.2 14.7 15.3agriculture 28.0 41.5 13.8 15.3

* class mark averages =2, standardized test scores=-1, highest parental education= primary or below, non-SEN,

non-Roma, non-repeater, current track is first choice, male

** class mark averages =5, standardized test scores=1, highest parental education= secondary or above, non-

SEN, non-Roma, non-repeater, current track is first choice, male

9. Robustness checks


25/36

apprenticeship training, since it controls for both local labor market effects as well as potential

differences between school qualities. Note that the HLCS has not used schools as sampling units, thus

the fact that some students are from the same school is chance only. There are 16 schools with only

one student in the sample. The average school has 3.7 students in the sample, which further

decreases to 3.1 if industry fixed effects are included. Taking missing values as well as the variance of

the outcome measure within school into account, and the fact that a representative subsample

should be used due to problems of weighting in fixed-effects logistic regressions, little less than 100

schools would be left for a non-linear analysis.16

Also since the multinomial logit model with a large

number of fixed effects has not yet been fully developed (see Pforr 2011), linear probability models

were estimated for this robustness check.17

The four category outcome of employed, unemployed,

study or other was transformed into a binary as in the propensity score matching test (employed vs.

everyone else).

The effect of apprenticeship training halves within schools, and loses its significance if school fixed-

effects are included in the linear model (Table 11). Apparently, the average apprenticeship student

does not have a greater chance of being employed than the average non-apprentice if they went to

the same school. However this effect is driven by the size of the training firm. Apprentices, who were

trained in small or in large firms, have exactly the same probability to be employed as non-


26/36

26

Table 11: Robustness check with industry and school fixed effects linear probability models

(1) (2) (3) (4) (5) (6) (7) (8)

apprentice, 4th

or 5th

wave 0.0986*** 0.127*** 0.0560 0.0570

(0.0380) (0.0464) (0.0477) (0.0629)

apprentice firm size, small (1-12) 0.0744 0.0993* -0.0117 0.0170

(0.0477) (0.0567) (0.0576) (0.0729)apprentice firm size, medium (13-100) 0.182*** 0.183*** 0.190*** 0.156*

(0.0518) (0.0675) (0.0629) (0.0877)

apprentice firm size, large (100+) 0.0559 0.120* 0.0131 0.0467

(0.0544) (0.0642) (0.0696) (0.0842)

Constant 0.365** 0.0342 0.442** 0.110 0.363** -0.0176 0.441** 0.0614

(0.164) (0.236) (0.183) (0.259) (0.166) (0.236) (0.186) (0.268)

Observations 964 675 957 670 936 666 929 661

R-squared 0.093 0.105 0.440 0.498 0.099 0.109 0.451 0.498

Industry FE no yes no yes no yes no yes

School FE no no yes yes no no yes yes

Dependent variable is employed=1, unemployed, studying or other =0, all variables as in the base model (Table 8 block 3) are controlled for

Robust standard errors in parentheses; *** p


27/36

The third robustness check sheds a bit of light on the firm-size puzzle. Figures 1 and 2 use another set

of outcome variables. The HLCS also asks students about their employment status during the last

academic year. That is, students in the 6th

wave of the study, in 2012 spring, were asked whether

they had had any regular job during the months between September 2010 (the start of the previous

school year) and August 2011, and students in the 5th

wave were asked whether they had a regular

job between September 2009 and August 2010. The data is for each month in between. Figure 1

depicts the marginal effect of apprenticeship training for a male, non-Roma, non-SEN student with

average class marks and test scores, parents with vocational education, who has not repeated class

up until 12th

grade, and applied for his track in the first place in 9th

grade, filled out the survey in May

2012 and have qualification from the average industry. The dependent variable is 1, if the student

had a regular job and 0 otherwise. Note that the outcome in May 2012 is the variable that was used

as in the estimations above.

It seems that apprentices are much more likely to find a regular job right after the end of the school

year. The marginal effect of apprenticeship training increases dramatically after the end of school

during the summer months, and declines afterwards. This indicates that apprentice VT students have

a quicker transition to the labor market than the non-apprentice VT students, but also that their

advantage slowly evaporates.18

The initial effect is also quite sizeable: it is around 19% in August


28/36

Figure 1. Marginal effect of apprentice students having a regular job

end of school-year 2010/11

end of school-year 2009/10

-.1

0

.1

.2

.3

10 Apr10 Jun

10 Sept10 Dec

11 Mar11 Jun

11 Aug12 May

95% confidence intervalls


29/36

Whether this effect is due to the superior specific skills that apprentices gained while being trained at

the mid-sized firms, or due to the screening of these firms, is not entirely clear from these figures.

Note however that the screening argument would predict an immediate and large difference

between the groupsbecause training firms hire the best candidates right after the trainingwhich

effect should converge by time, as firms hire new employees. The human capital argument, on the

other hand, would suggest a steady but continuous increase in the gap, which should only fade away

after a good amount of time, when others also gain the specific skills necessary for employment. The

figures, thus, support the screening rather than the human capital arguments. Also it is likely that

small firms rely less on institutionalized screening processes, as they are more likely to use their

social networks to recruit apprentices and thus rely less on this probation period and more on

other information channels (e.g. take relatives or friends as apprentices in the first place).

Conversely, medium or large firms are more likely to have institutionalized mechanisms for selecting

apprentices and new employees. Moreover, turnover at a large firm is probably larger, at least in

sheer numbers, which suggests that the potential advantage that an apprentice can gain from being

selected early diminishes quickly as the firm hires new workforce. Nevertheless this analysis in itself

is not enough to see whether the screening or the human capital argument comes closer to reality,

19


30/36

they finish education, than those who had their practical training in schools and are otherwise similar

to the workplace-based group. The effect is net of individual skills, school attainment, parental

background, motivation, gender and ethnicity, and robust to the inclusion of industry fixed effects,

and for school fixed effects but only for students trained in mid-sized firms. Also results show that

the significant marginal effect of apprenticeship training declines rapidly for students trained at large

firms, while this decline is less marked in medium or small firm trained apprentices, suggesting that

large turnover could eliminate the positive effects of apprenticeship training more quickly.


31/36

References

Baranowska, Anna, Michael Gebel, and Irena E. Kotowska. 2011. The Role of Fixed-term Contracts atLabour Market Entry in Poland: Stepping Stones, Screening Devices, Traps or Search

Subsidies? Work, Employment & Society25 (4) (December 1): 777793.doi:10.1177/0950017011419705.

Bertschy, Kathrin, M. Alejandra Cattaneo, and Stefan C. Wolter. 2009. PISA and the Transition intothe Labour Market. LABOUR23: 111137. doi:10.1111/j.1467-9914.2008.00432.x.

Bol, Thijs, and Herman G. van de Werfhorst. 2011. Signals and Closure by Degrees: The EducationEffect Across 15 European Countries. Research in Social Stratification and Mobility29 (1)

(January): 119132. doi:10.1016/j.rssm.2010.12.002.Bonnal, Liliane, Sylvie Mendes, and Catherine Sofer. 2002. School-to-Work Transition:

Apprenticeship Versus Vocational School in France. International Journal of Manpower23(5): 426442. doi:10.1108/01437720210436046.

Breen, Richard. 2005. Explaining Cross-National Variation in Youth Unemployment: Market andInstitutional Factors. European Sociological Review21 (2): 125134.

CEDEFOP. 2011. Vocational Education and Training in Poland. Luxembourg: Publications Office ofthe European Union.

Eichhorst, Werner, Nria Rodrguez-Planas, Ricarda Schmidl, and Klaus F. Zimmermann. 2012. ARoadmap to Vocational Education and Training Systems Around the World. IZA DP(7110):46.

Horn, Dniel. 2010. The Political Background of Structural Changes in the Educational System ofHungary, 1985-1994. Center for Policy Studies Working Paper(2010/1): 45.

. 2013a. Diverging Performances: The Detrimental Effects of Early Educational Selection onEquality of Opportunity in Hungary. Research in Social Stratification and Mobility.


32/36

Pforr, Klaus. 2011. Implementation of a Multinomial Logit Model with Fixed Effects. German StataUsers Group Meetings 2011. Stata Users Group.

http://econpapers.repec.org/paper/bocdsug11/03.htm.Piopiunik, Marc, and Paul Ryan. 2012. Improving the Transition Between Education/training and theLabour Market: What Can We Learn from Various National Approaches? EENEE AnalyticalReport(13.).

Plug, Erik, and Wim Groot. 1998. Apprenticeship Versus Vocational Education: Exemplified by theDutch Situation. Unpublished Manuscript.

Rosenbaum, James E, Takehiko Kariya, Rick Settersten, and Tony Maier. 1990. Market and NetworkTheories of the Transition from High School to Work: Their Application to Industrialized

Societies.Annual Review of Sociology16: 263299.Ryan, Paul. 2001. The School-to-Work Transition: A Cross-National Perspective.Journal of

Economic Literature39 (1): 3492.Shavit, Yossi, and Walter Mller. 2000. Vocational Secondary Education: Where Diversion and

Where Safety Net? European Societies21 (1): 2950.Van de Werfhorst, Herman G. 2011a. Skills, Positional Good or Social Closure? The Role of Education

Across Structuralinstitutional Labour Market Settings.Journal of Education and Work24(5): 521548. doi:10.1080/13639080.2011.586994.

. 2011b. Skill and Education Effects on Earnings in 18 Countries: The Role of NationalEducational Institutions. Social Science Research40 (4) (July): 10781090.doi:10.1016/j.ssresearch.2011.03.004.

West, John. 2013. Vocational Education and Training in Eastern Europe: Transition and Influence.LLAKES Research Paper(41): 68.

Wolbers, Maarten H. J. 2007. Patterns of Labour MarketEntry: A Comparative Perspective onSchool-to-Work Transitions in 11 European Countries.Acta Sociologica50 (3): 189210.

Wolter, Stefan C., and Paul Ryan. 2011. Apprenticeship. In , edited by Stephen Machin Eric A.


33/36

Appendix A

Figure A.1 The Hungarian compulsory education system

HUNGARY 2009/2010

level ISCED 0 ISCED 1 ISCED 2 ISCED 3

1st cycle 2nd cycle

Grade 1 2 3 4 5 6 7 8 9 10 11 12 13

kindergarten general school academic secondary school prog. ISCED 3a

+

vocational secondary school prog. ISCED 3a

(technikum)

+

vocational training prog. ISCED 3c ++

Age 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

compulsory education until the age of 18 applies for the 1st graders in 1998 and later (previously and from September 2012: until the age of

16)

vocational secondary school programs curriculum includes vocational subjects and many students progress to PS voc to get a VQ

+ : some schools offer an extra grade teaching a foreign language before secondary school educ. (i.e. between grade 8 and 9)

++: some programs are also available for elementary school drop-outs

ISCED English national language share

0 kindergarten voda

1,2a general school ltalnos iskola 100%

3a academic secondary school prog. gimnzium

3a vocational secondary school prog. szakkzpiskola

3c vocational training prog. szakiskola


34/36

Table A1: Old and new categories of the national training register (OKJ)

New categories (industries) Original categories in the national training register

Social Services HealthSocial services

Education

Art, culture, communication

Mechanics EngineeringElectrical-engineering, electronics

Informatics

Industry Chemical industry

ArchitectureLight industry

Wood industry

Printing industry

Transportation-environment TransportationEnvironment and water-management

Services Business and economicsManagement

Trade, marketing and administration

Catering, tourismOther Services

Agriculture AgricultureFood industry


35/36

35

Table A2: Types of training and employer firms and the number of individuals

1 2 3 4 5 6 7 8 9 1

0

1

1

1

2

1

3

1

4

15 1

6

1

7

1

8

19 2

0

2

1

2

2

2

3

mis total

5thwavetr

ainer

1 Agriculture, forestry and fishing 1 0 1 0 1 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 12 17

2 Mining and quarrying 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1

3 Processing 6 0 29 2 0 2 1 0 2 0 0 0 0 0 1 1 0 0 2 0 0 0 0 56 102

4 Electricity, gas, steam and air conditioning 1 0 0 7 1 2 0 1 0 0 0 0 0 0 2 0 0 0 1 0 0 0 1 14 30

5 Water supply, wastewater collection and

treatment, waste management

1 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 5

6 Construction 10 0 17 1 1 45 1 1 1 1 0 0 1 1 4 0 1 0 5 0 0 0 0 81 171

7 Trade, automotive services 1 0 15 0 0 4 20 2 7 0 0 0 0 0 3 0 0 0 8 0 0 0 0 70 130

8 Transportation, warehousing 0 0 3 0 0 0 3 1 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 7 16

9 Hotels and restaurants, catreing 2 0 10 0 0 1 17 0 64 0 0 0 0 0 0 0 1 2 5 0 0 0 0 129 231

10 Information, communication

11 Financial and insurance activities 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 4 4

12 Real estate transactions

13 Professional, scientific and technical activities 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1

14 Administrative and support service activities 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 2

15 Administration and defense, compulsory social

security

0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 0 0 0 0 0 1 4

16 Education 0 0 3 0 0 0 2 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 9 15

17 Human health and social work 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 21 24

18 Arts, entertainment and recreation 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 1 3

19 Other services 2 0 7 0 3 3 5 2 8 0 0 0 0 1 3 0 1 0 19 0 0 0 1 70 125

20 Households as employers, producers, and service 0 0 1 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3

21 Organizations outside Hungary

22 Other

23 Don't know 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 0 0 0 6 8

missing 21

1

2 44

3

2

4

4

6

15

4

34

5

7

1

21

3

3

3

4

0

1 1

2

2

5

15

4

2

3

5

4

5

8

22

6

1

1

1 4 1

2

6,9

67

9,13

total 23

5

2 53

3

3

4

5

2

21

3

39

5

7

8

29

6

3

4

4

0

1 1

3

2

7

17

3

2

5

5

9

6

0

26

9

1

1

1 4 1

4

7,4

53

10,0

22


36/36

workplace-based vocational training in central and eastern europe - evidence on the school to work...

Documents